Can I use ChatGPT (or any other AI) to represent me at the FTT?

How has AI fared too date?

There have been several cases where individuals representing themselves have submitted case law generated by AI that didn’t appear in well-known case law archive databases such as the British and Irish Legal Information Institute (BAILII – https://www. bailii.org/) and the National Archives – https://caselaw.nationalarchives.gov.uk/

Examples include:

Napier House v Assethold (Jul 2025): There were ten fake or irrelevant citations in the respondent’s appeal grounds, including one about Parole Board hearings that led, in part, to an appeal being dismissed.

Holloway v Beckles (Aug 2025): A litigant-in-person leaseholder cited three nonexistent cases. The FTT treated this as unintentional, but a Rule 13(b) costs order of £750 followed.

…and it’s not just amateurs that get caught out.

In Victoria Place Flats v Assethold (Oct 2025), a solicitor respondent relied on a wrongly applied case plus AI-generated fake citations from Copilot; the judge personally tested the AI tool, found it consistently wrong, and flagged the solicitor’s conduct to the Regional Tribunal Judge.

The website https://naturalandartificiallaw.com/ has reported that, as of July 2026, there have been more than 1,700 instances of AI hallucinations in legal proceedings worldwide, and 67 in the UK alone.

So why does this happen?

1. AI models don’t “think” in the human context
In simplistic terms, they are super-advanced pattern recognition machines. They predict plausible text, not verified facts. Large Language Models (LLM’s) generate each next word based on statistical patterns learned from training data — it’s optimising for “what sounds like a right answer” rather than checking a database of real judgments.

2. Case citations have a very learnable format
UK case law follows rigid patterns — party names, “v”, year, court abbreviation, neutral citation number, paragraph references. AI models have seen thousands of real examples of that pattern, so they can generate a citation that’s structurally perfect and stylistically convincing while being entirely invented, because it’s reproducing the shape of a citation, not retrieving an actual case.

3. Base LLM’s don’t look anything up when answering.
AI’s generate responses from memory of patterns in training data. (unless the AI is specifically using a search or legal-database tool). It has no internal mechanism that says “does this case actually exist” or any internal signal distinguishing between “I’m certain of this” from “I’m making this up”…..but it will deliver its output with tremendous confidence.

4. Sparse or overlapping training data gets blended
Specific, narrow legal questions may have had few real cases directly on point in training data (see example below) The model can effectively merge fragments of several tangentially related cases and legal principles into one fluent but fictional “authority” — which is exactly what happened in Victoria Place Flats v Assethold.

5. Answers aren’t stable
As the tribunal judge in Victoria Place found, asking the same question later or with a punctuation tweak produced different fake citations. Outputs are probabilistic rather than fixed facts, so each response is newly generated and may change as learning progresses.

Common Ground and AI

At Common Ground we were very early adopters of AI, initially using it to read financial data from invoices directly into our accounts system. We also use AI to generate letters, analyse leases, conduct research (but not write Leasehold Library articles), analyse data and so on. More importantly, what AI doesn’t do at Common Ground is:

1. Answer the phone. We have lovely humans for that.

2. Replace staff.

Point number two is extremely important and covered here: https://www.commongroundestates.co.uk/leasehold-library/will-artificial-intelligencereplace-property-managers/

Our experience of some of the pitfalls of AI is as follows:

• When analysing leases, you absolutely must include the whole lease and not just excerpts. Leases are written as holistic documents, not just a series of legal requirements.

• Even when analysing leases, if the AI doesn’t understand part of a lease, it may simply ignore it.

• It may not make the link to relevant legislation when interpreting leases. By way of example, I have particular expertise in retrofitting EV chargers into communal environments, as well as being a qualified property manager. I asked Copilot whether the service charge could be used to pay for the installation of a communal EV system. Initially, the AI relied on a sweeper clause to confidently conclude that it could. I interrogated it further using legal opinion obtained from other sites where I have installed EV chargers, and it quickly changed its mind.

• Other documents may be required to form an opinion. For example, this may be the Memorandum and Articles of a residents’ management company in conjunction with a lease. An AI may utilise standard company law and standard formats in lieu of these documents, but a company’s Memorandum and Articles can be customised, so the context is important.

Conclusions

By all means use AI to understand your rights as a leaseholder, but be extremely careful about using it to represent yourself at the FTT.

At Common Ground I tell my staff to predict the answer they believe the AI should generate, and to interrogate it further if that differs from the expected outcome. In other words, possessing the knowledge and skills to interrogate AI properly is fundamental to achieving a good outcome.

Alan Draper, FTPI AssocRICS, Managing Director, CommonGround Estates